← Back to newsletters

AI Daily Brief — Sun May 24

2026-05-24

By Vadym · Generated with AI, curated by me


Listen to this issue

A quiet Sunday with a loud week underneath it. AI proved an 80-year-old theorem, a VS Code extension robbed GitHub, and every Copilot user is about to get a new kind of bill.


Headlines & News
Research

OpenAI Reasoning Model Autonomously Disproves 80-Year-Old Erdős Conjecture

An internal OpenAI reasoning model independently disproved a central conjecture in discrete geometry first posed by Paul Erdős in 1946 — the planar unit distance problem, which concerns the maximum number of point pairs at unit distance in a plane. For nearly 80 years, mathematicians believed square grids were optimal. The model found an infinite family of constructions using deep algebraic number theory that beats the grid, with the bound later refined by Princeton mathematician Will Sawin. Fields medalist Tim Gowers described it as “a milestone in AI mathematics.” Companion remarks from multiple mathematicians confirmed independent verification of the proof.

Source: OpenAI

Security

Poisoned Nx Console VS Code Extension Breached GitHub’s Internal Repositories

A compromised version of the Nx Console VS Code extension (v18.95.0) was published to the VS Code Marketplace on May 18 and stayed live for approximately 11 minutes before the Nx team pulled it. The extension has 2.2 million installations. Within seconds of a developer opening any workspace, the malicious version silently fetched an obfuscated payload that harvested credentials from GitHub, npm, AWS, HashiCorp Vault, and Kubernetes, then exfiltrated them via HTTPS, the GitHub API, and DNS tunneling. Roughly 3,800 GitHub internal repositories were accessed. The threat group TeamPCP claimed responsibility; root cause traced to a prior supply chain attack against TanStack that leaked an Nx developer’s GitHub credentials.

Source: The Hacker News

Coding

GitHub Copilot Switches to Usage-Based Billing on June 1 as Developers Push Back

All GitHub Copilot plans transition to GitHub AI Credits on June 1, replacing the previous Premium Request Units system. Every plan gets a base credit allotment plus a flex pool that scales with usage — meaning costs can exceed the fixed subscription price. Developers have been vocal: the old request-based model was imperfect, but predictable; the new token-based system is technically fairer but harder to reason about before the bill arrives. Many are asking for clearer per-model token costs and rollover credit policies. Base subscription prices are not changing — only the ceiling is lifting.

Source: GitHub Blog

Robotics

August Robotics Raises $30M to Deploy Autonomous Drilling Robots at AI Data Centers

August Robotics closed a $30 million Series B led by Big Pi Ventures, with participation from Blackbird, Skip Capital, and GS Futures. The company builds autonomous drilling robots developed with DeWalt, already operating at construction sites across the US and Europe — including active deployments at AI data center builds. Funding will support AI development, new product lines, and expansion into Europe and Australia. The detail worth flagging: the physical infrastructure buildout for AI is itself increasingly being executed by robots. That loop is now live, not planned.

Source: The AI Insider

Models

Gemini 3.5 Flash Goes GA: 4x Faster, 40% Cheaper, Benchmarks Past Frontier Rivals on Coding

Google shipped Gemini 3.5 Flash generally available at Google I/O 2026. The model runs 4x faster than comparable frontier models, costs $1.50/$9 per million input/output tokens (with a 90% discount on cached input), and ships with a 1M-token context window and full multimodal support. It outperforms Gemini 3.1 Pro on coding and agentic benchmarks while being significantly cheaper to operate. Google’s stated bet is that the next AI wave is agents, not chatbots — and Flash is priced to make that workload sustainable at scale. Every company building agentic pipelines should be re-running cost models today.

Source: TechCrunch


Analysis

Takeaway

This week handed us three separate signals about what AI is actually becoming. OpenAI’s math proof is the clearest evidence yet that reasoning models can do genuinely novel intellectual work — not assistance, not retrieval, but original contribution. The Nx Console attack is a reminder that as developer tooling becomes the core of AI infrastructure, it also becomes the highest-value attack surface on the internet. And Copilot’s billing change is where the rubber meets the road for enterprise adoption: the “unlimited AI” era is ending, replaced by metered costs that organizations will have to justify line by line. Math, security, and money. All three are load-bearing now.

— Boba


Curated by Vadym